Blind motion deblurring using image statistics
Transcript of Blind motion deblurring using image statistics
Blind motion deblurringusing image statistics
Supplementary Material
Anat LevinSchool of Computer Science and Engineering
The Hebrew University of Jerusalem
Example 1-input
Example 1- deblurring the entire image
12 tap kernel
Example 1- result
Example 1-recomparing to input
Vertical edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk12 pixels blur(i)
Gray: lkunblurred(i) < lk12 pixels blur(i)
Example 1- local evidence
Example 1- inferred segmentation
Example 2-input
Example 2- deblurring the entire image
4 tap kernel
Example 2- result
Example 2- recomparing to input
Example 2- local evidence
Vertical edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk4 pixels blur(i)
Gray: lkunblurred(i) < lk4 pixels blur(i)
Example 2- inferred segmentation
Example 2 with wrong histograms -input
Example 2 with wrong histograms -deblurring the entire image
6 tap kernel
Example 2 with wrong histograms -result
Example 2 with wrong histograms –recomparing to input
Example 2 with wrong histograms -local evidence
Vertical edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk6 pixels blur(i)
Gray: lkunblurred(i) < lk6 pixels blur(i)
Example 2 with wrong histograms -inferred segmentation
Example 3- input
In orig size
zoomed
Example 3- deblurring the entire image
In orig size
zoomed
6 tap kernel
Example 3- result
In orig size
zoomed
Example 3- recomparing to input
In orig size
zoomed
Example 3- local evidence
Vertical edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk6 pixels blur(i)
Gray: lkunblurred(i) < lk6 pixels blur(i)
Example 3- inferred segmentation
Example 4 (extracting 3 layers) -input
Example 4 (extracting 3 layers) -deblurring the entire image
1st kernel- 2 tap
Example 4 (extracting 3 layers) -deblurring the entire image
2nd kernel- 9 tap
Example 4 (extracting 3 layers)-result
Example 4 (extracting 3 layers) -recomparing to input
Example 4 (extracting 3 layers) -local evidence
Vertical edges map and the maximum likelihood model in each pixel
White: unblurred
Light Gray: 2 pixels blur
Dark gray: 9 pixels blur
Example 4 (extracting 3 layers) -inferred segmentation
Example 5 (extracting 3 layers) -input
In orig size
zoomed
Example 5 (extracting 3 layers) -deblurring the entire image
1nd kernel- 4 tap
In orig size
zoomed
Example 5 (extracting 3 layers) -deblurring the entire image
2nd kernel- 8 tap
In orig size
zoomed
Example 5 (extracting 3 layers)-result
In orig size
zoomed
Example 5 (extracting 3 layers) -recomparing to input
In orig size
zoomed
Example 5 (extracting 3 layers) -local evidence
Vertical edges map and the maximum likelihood model in each pixel
White: unblurred
Light Gray: 4 pixels blur
Dark gray: 8 pixels blur
Example 5 (extracting 3 layers) -inferred segmentation
Example 6 (non horizontal blur)-input
Example 6 (non horizontal blur)-estimated blur direction
Example 6 (non horizontal blur)-deblurring the entire image
26 tap kernel
Example 6 (non horizontal blur)-result
Example 6 (non horizontal blur)-recomparing to input
Edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk26 pixels blur(i)
Gray: lkunblurred(i) < lk26 pixels blur(i)
Example 6 (non horizontal blur)-local evidence
Example 6 (non horizontal blur) -inferred segmentation
Example 7 (non horizontal blur)- input
Example 7 (non horizontal blur)-estimated blur direction
Example 7 (non horizontal blur)-deblurring the entire image
15 tap kernel
Example 7 (non horizontal blur)- result
Example 7 (non horizontal blur)-recomparing to input
Edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk15 pixels blur(i)
Gray: lkunblurred(i) < lk15 pixels blur(i)
Example 7 (non horizontal blur)- local evidence
Example 7 (non horizontal blur)- inferred segmentation
Failure example - input
Failure example - deblurring the entire image
6 tap kernel
Failure example - result
Failure example – recomparing to input
Edges map and the maximum likelihood model in each pixel
White: lkunblurred(i) > lk6 pixels blur(i)
Gray: lkunblurred(i) < lk6 pixels blur(i)
Failure example - local evidence
Failure example - inferred segmentation
Comparison- using unsupervised segmentation
input
Comparison- using unsupervised segmentation
segments + sizes of fitted blur model
13
1
1
3
18
Comparison- using unsupervised segmentation
result
Comparison- using unsupervised segmentation
recomparing to input
Comparison- using unsupervised segmentation
recomparing to our result